I am working with mark-recapture models in MARK. I have my survival estimates (best model included time and group parameters). I want to statistically compare my estimates between groups as they seem to be really similar (despite the group effect). I am not sure about which statistical test/appproach to use.
Mostly this sort of study does not have replicates, multiple estimates from the same population using samples sufficiently separated that the samples are independent. If you do have replicates then something as simple as a t-test could work.
The other choice is to randomize individuals between groups. This could work if you have many individuals: at least 20 in each group and preferably many more. If there are no differences between groups then it does not make any difference which group an individual came from. The problem will be to link the randomization method with MARK. I do not know how to do this, or even if it can be done. If there is a package in R that can do this sort of analysis (something like https://cran.r-project.org/web/packages/Rcapture/Rcapture.pdf) then this method would be easy (if you already know R).
Hi Juan,
How many groups do you have? Are the beta values for each group significant (ie., the 95% CI for the beta values do not include zero)? If not, this may be why some of your estimates among groups are similar. If a model with group effects is favored over a model with no group effects that doesn’t necessarily mean that all groups are different from one another. To investigate this, as an example, let’s assume you have four groups. You may want to compare your model with four groups to a model with three groups (by having the two most similar groups share the same slope) or two groups (by pairing the two most similar groups with each other). Then compare these models using the standard methods for multimodel inference and model selection (e.g., AICc). Alternatively, because the models I described are nested, you could compare them using a likelihood ratio test instead. Just keep in mind that AIC ranking operates on what translates to roughly 85% CI (i.e., p ~0.15), whereas p values from likelihood ratio tests are based on 95% CI (p=0.05).
Hope this helps. Cheers!
Hi Juan,
You can also use the Wilcoxon signed-rank test that is a non-parametric test and allows to compare paired observations from two samples. Thus you can apply this test in R using the function wilcox.test( ).
Best redards!
- Badges
- Science method
- Similar topics
- Biological Science
- Ecology